KernelAbstractions: Use fine-grained synchronization (#274)

JuliaGPU · Jan 10, 2025 · dc8eef9 · dc8eef9
1 parent 9034a6a
commit dc8eef9
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/src/OpenCLKernels.jl b/src/OpenCLKernels.jl
@@ -22,7 +22,7 @@ KA.zeros(::OpenCLBackend, ::Type{T}, dims::Tuple) where T = OpenCL.zeros(T, dims
 KA.ones(::OpenCLBackend, ::Type{T}, dims::Tuple) where T = OpenCL.ones(T, dims)
 
 KA.get_backend(::CLArray) = OpenCLBackend()
-KA.synchronize(::OpenCLBackend) = cl.device_synchronize()
+KA.synchronize(::OpenCLBackend) = cl.finish(cl.queue())
 KA.supports_float64(::OpenCLBackend) = false  # XXX: this is platform/device dependent
 
 Adapt.adapt_storage(::OpenCLBackend, a::Array) = Adapt.adapt(CLArray, a)